Descripción de negocio

La Clínica de los Alpes desea conocer cuales son los principales factores de muerte que disminuyen la expectativa de vida de los ciudadanos de su país para realizar campañas de concientización con el fin de mejorar la calidad de vida de los ciudadanos. La Clínica cuenta con un conjunto de datos sobre las expectativas de vida a lo largo de los años en los Alpes y países cercanos, junto con algunos indicadores que miden la salud de una persona, como lo puede ser el índice de masa corporal, la incidencia de varias enfermedades y algunos factores socioculturales como el consumo de alcohol o tabaco. Esta información quieren utilizarla para construir un modelo que les pueda ayudar para resolver las siguientes tareas:

Identificar las variables que más impactan en la expectativa de vida de la gente de los Alpes. Predecir la expectativa de vida en los Alpes a partir de las variables de interés.

1. Carga de librerias

seed = 161
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from pandas_profiling import ProfileReport


# Composicion de pipelines
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import PolynomialFeatures, Normalizer
from sklearn.preprocessing import MinMaxScaler,StandardScaler, MaxAbsScaler
from sklearn.model_selection import train_test_split

# Regresion lineal
from sklearn.linear_model import LinearRegression

# Importar/ Exportar modelos
from joblib import dump, load
from itertools import combinations

# Metricas
from sklearn.metrics import mean_squared_error as mse

# q-q plots
import scipy.stats as stats

# Visualizacion de datos
%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns

sns.set_palette("Paired")

2. Carga y limpieza de los datos

data_or = pd.read_csv('datos.csv', na_values="NA-VALUE")
print(data_or.shape)
data_or.head(5)

(294, 19)

	Expectancy	Adult Mortality	infant deaths	Alcohol	percentage expenditure	Hepatitis B	Measles	BMI	under-five deaths	Polio	Total expenditure	Diphtheria	HIV/AIDS	GDP	Population	thinness 10-19 years	thinness 5-9 years	Income composition of resources	Schooling
0	2644	151.0	0	1.80	423.295351	9.0	0	68.6	0	91.0	4.87	9.0	0.1	2284.378580	146.0	0.1	0.1	0.693	14.6
1	2645	153.0	0	1.79	45.851058	85.0	0	67.8	0	91.0	5.90	9.0	0.1	229.714718	99789.0	0.1	0.1	0.683	13.7
2	2646	155.0	0	1.51	310.820338	88.0	0	67.0	0	85.0	5.30	84.0	0.1	1842.444210	99184.0	0.1	0.1	0.679	13.5
3	2647	157.0	0	1.35	330.100739	91.0	4	66.2	0	91.0	5.66	89.0	0.1	1837.977391	98611.0	0.1	0.1	0.674	13.2
4	2648	158.0	0	1.24	40.491289	93.0	0	65.5	0	91.0	4.75	91.0	0.1	263.272360	9882.0	0.1	0.1	0.676	13.7

sns.pairplot(data_or,height=3, y_vars='Expectancy',x_vars =data_or.columns[1:7])
sns.pairplot(data_or,height=3, y_vars='Expectancy',x_vars =data_or.columns[7:13])
sns.pairplot(data_or,height=3, y_vars='Expectancy',x_vars =data_or.columns[13:19])

<seaborn.axisgrid.PairGrid at 0x27a9bd9a280>

data_or=data_or.dropna(axis=0)
#data_or.drop(data_or.loc[data_or['Adult Mortality']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['infant deaths']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Alcohol']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['percentage expenditure']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Hepatitis B']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Measles']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['BMI']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['under-five deaths']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Polio']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Total expenditure']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Diphtheria']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['HIV/AIDS']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['GDP']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['thinness  10-19 years']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['thinness 5-9 years']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Income composition of resources']==0].index, inplace=True)
#data_or.drop(data_or.loc[data_or['Schooling']==0].index, inplace=True)

#data_or['Adult Mortality']=data_or['Adult Mortality' !=0]
#data_or['infant deaths']=data_or['infant deaths'!=0]
#data_or['Alcohol']=data_or['Alcohol'!=0]
#data_or['percentage expenditure']=data_or['percentage expenditure'!=0]
#data_or['Hepatitis B']=data_or['Hepatitis B'!=0]
#data_or['Measles']=data_or['Measles'!=0]
#data_or['BMI']=data_or['BMI'!=0]
#data_or['under-five deaths']=data_or['under-five deaths'!=0]
#data_or['Polio']=data_or['Polio'!=0]
#data_or['Total expenditure']=data_or['Total expenditure'!=0]
#data_or['Diphtheria']=data_or['Diphtheria'!=0]
#data_or['HIV/AIDS']=data_or['HIV/AIDS'!=0]
#data_or['GDP']=data_or['GDP'!=0]
#data_or['thinness  10-19 years']=data_or['thinness  10-19 years'!=0]
#data_or['thinness 5-9 years']=data_or['thinness 5-9 years'!=0]
#data_or['Income composition of resources']=data_or['Income composition of resources'!=0]
#data_or['Schooling']=data_or['Schooling'!=0]

#'Adult Mortality','infant deaths','Alcohol','percentage expenditure','Hepatitis B','Measles','BMI','under-five deaths',
#'Polio','Total expenditure','Diphtheria','HIV/AIDS','GDP','Population','thinness  10-19 years','thinness 5-9 years',
#'Income composition of resources','Schooling'
data_or.describe()

	Expectancy	Adult Mortality	infant deaths	Alcohol	percentage expenditure	Hepatitis B	Measles	BMI	under-five deaths	Polio	Total expenditure	Diphtheria	HIV/AIDS	GDP	Population	thinness 10-19 years	thinness 5-9 years	Income composition of resources	Schooling
count	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	294.000000	2.940000e+02	294.000000	294.000000	294.000000	294.000000
mean	2790.500000	180.156463	22.748299	4.031327	250.691789	67.258503	2299.707483	39.811565	31.921769	82.459184	5.934014	80.578231	2.866327	2888.804225	4.541904e+06	5.141497	5.135374	0.492966	9.931293
std	85.014705	149.969676	28.065706	3.411991	636.324313	35.669719	6887.681389	20.323780	43.125549	21.932024	3.285364	24.922111	6.876873	7269.383426	1.293499e+07	4.007686	4.123723	0.289985	4.827973
min	2644.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	1.000000	0.000000	7.000000	0.000000	5.000000	0.100000	0.000000	0.000000e+00	0.100000	0.100000	0.000000	0.000000
25%	2717.250000	82.000000	3.000000	1.375000	0.000000	53.000000	0.000000	19.300000	3.000000	75.000000	4.400000	75.000000	0.100000	0.000000	0.000000e+00	1.625000	1.600000	0.415000	9.725000
50%	2790.500000	153.000000	10.000000	2.720000	27.137321	83.000000	55.500000	43.000000	12.000000	92.000000	5.405000	92.000000	0.100000	430.824070	1.055190e+05	5.050000	4.900000	0.600500	11.100000
75%	2863.750000	231.000000	29.000000	6.632500	194.536691	94.000000	816.500000	57.475000	42.000000	96.000000	7.075000	96.000000	0.775000	2244.678564	2.482152e+06	6.675000	6.700000	0.715750	12.975000
max	2937.000000	723.000000	116.000000	12.220000	4003.908598	99.000000	49871.000000	79.300000	191.000000	99.000000	17.600000	99.000000	43.500000	45758.955400	7.827147e+07	15.800000	16.400000	0.836000	15.700000

restr = data_or.apply(lambda x:np.abs(stats.zscore(x))<3).all(axis=1)
data_cl = data_or.drop(data_or.index[~restr],inplace=False)
data_cl.shape

(230, 19)

sns.pairplot(data_cl,height=3, y_vars='Expectancy',x_vars =data_or.columns[1:7])
sns.pairplot(data_cl,height=3, y_vars='Expectancy',x_vars =data_or.columns[7:13])
sns.pairplot(data_cl,height=3, y_vars='Expectancy',x_vars =data_or.columns[13:19])

<seaborn.axisgrid.PairGrid at 0x27a99641f70>

ProfileReport(data_cl)

sns.distplot(data_cl['Expectancy'])

C:\Users\NikCa\anaconda3\lib\site-packages\seaborn\distributions.py:2619: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).
  warnings.warn(msg, FutureWarning)

<AxesSubplot:xlabel='Expectancy', ylabel='Density'>

# Se selecciona la variable objetivo, en este caso "Adult Mortality".
Y = data_cl['Expectancy']
# Del conjunto de datos se elimina la variable "Adult Mortality"
X = data_cl.drop(['Expectancy'], axis=1)

X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=seed)
df_train = pd.concat([X_train,Y_train],axis=1)
df_test =pd.concat([X_test,Y_test],axis=1)

Colinealidad

#'Adult Mortality','infant deaths','Alcohol','percentage expenditure','Hepatitis B','Measles','BMI','under-five deaths',
#'Polio','Total expenditure','Diphtheria','HIV/AIDS','GDP','Population','thinness  10-19 years','thinness 5-9 years',
#'Income composition of resources','Schooling'

selected_cols =(
    'Adult Mortality','infant deaths','Alcohol','percentage expenditure','Hepatitis B','BMI',
'Polio','Total expenditure','Diphtheria'
    
)
sns.heatmap(pd.DataFrame(X_train, columns= selected_cols).corr(method='kendall'),center=0, )

<AxesSubplot:>

pre = [
       ('initial',ColumnTransformer([("selector", 'passthrough',selected_cols)])),
       ('scaler',MinMaxScaler()),
       ('norm',Normalizer()),

]
model = [('model', LinearRegression())]
pipeline = Pipeline(pre+model)
pipeline = pipeline.fit(X_train,Y_train)
print(pd.DataFrame({'columns':selected_cols,'coef':pipeline['model'].coef_}))
pipeline.score(X_test,Y_test)

                  columns        coef
0         Adult Mortality  241.402131
1           infant deaths  -39.209483
2                 Alcohol  -53.037230
3  percentage expenditure -247.890629
4             Hepatitis B   11.430407
5                     BMI  -26.487210
6                   Polio -111.405566
7       Total expenditure  191.696327
8              Diphtheria  103.032329

0.39992860253177465

Root-Mean-Square Error (RMSE)

# Note que hay que sacarle la raiz al valor
np.sqrt(mse(Y_test, pipeline.predict(X_test)))

62.64274117642234

Normalidad en los Errores

# Calculamos los errores
errors = (pipeline.predict(X_test) - Y_test).values

fig, axes = plt.subplots(1, 2, figsize = (12,4))

# Dispersión
sns.scatterplot(x = pipeline.predict(X_test), y = errors, alpha = 0.5, ax = axes[0])

# q-q plot
_ = stats.probplot(errors, dist="norm", plot=axes[1])

Homocedasticidad

sns.scatterplot(data  = df_test, x = 'Adult Mortality', y = errors, alpha = 0.2)

<AxesSubplot:xlabel='Adult Mortality'>

Exportar Modelo

filename = 'modelo.joblib'
dump(pipeline,filename)

['modelo.joblib']